Prediction of groundwater quality is an essential step for sustainable utilization of water resources. Most of the related research in the study area focuses on water distribution and rational utilization of resources but lacks results on groundwater quality prediction. Therefore, this paper introduces a prediction model of groundwater quality based on a long short-term memory (LSTM) neural network. Based on groundwater monitoring data from October 2000 to October 2014, five indicators were screened as research objects: TDS, fluoride, nitrate, phosphate, and metasilicate. Considering the seasonality of water quality time series data, the LSTM neural network model was used to predict the groundwater index concentrations in the dry and rainy periods. The results suggest the model has high accuracy and can be used to predict groundwater quality. The mean absolute errors (MAEs) of these parameters are, respectively, 0.21, 0.20, 0.17, 0.17, and 0.20. The root mean square errors (RMSEs) are 0.31, 0.29, 0.28, 0.27, and 0.31, respectively. People can be given early warnings and take measures according to the forecast situation. It provides a reference for groundwater management and sustainable utilization in the study area in the future and also provides a new idea for coastal cities with similar hydrogeological conditions.

Architectural construction is responsible for the consumption of large amounts of resources, so the optimization of architectural design and evaluation is significant for sustainable global development. Most architectural assessments focus on energy conservation, novel materials and eco-friendly strategies, but without agreed indicators and criteria. Since the consideration of natural aspects is somewhat fuzzy and vague, this study utilized data mining technology to explore the major factors related to relationships between buildings and nature. By employing the popular technique of web crawling, this study collected 38,320 architectural descriptions from the “Archdaily”, including descriptions of 11 types of buildings, four of which were taken as typical research representatives. The 100 most frequent words were used to create a word cloud. Using Python script, all of the text was refined and processed with the word2vec model, thereby allowing to conduct Agglomerative Hierarchical Clustering (AHC). The frequency of words related to natural aspects were analyzed within 15 architectural design elements. Different building types in different areas have obvious similarities in terms of design elements, so it is feasible to adopt the same evaluation factors for the building evaluation systems of different regions. This paper mainly focuses on improving the accuracy and validity of assessment by providing basic evaluation indicators that could enhance connections between design and evaluation progress, stimulating the improvement of building environmental performance.